Some papers in Image Analysis


Homeostatic image perception: An artificial system, T. Feldman and L. Younes, Computer Vision and Image Understanding  102  70--80  (2006)

In this paper, we build a low-level "vision system" in successive tiers. The image is transformed into a series or ternary fields that collect coarse local information, only storing whether the image response to a local filter is lower than a high threshold, or smaller than a low threshold, or neither (first image). A Markov random field is then trained to learn a join distribution for these layers, within a finite field of view(second image). This model is then used as a saliency detector, selecting image patches that are atypical with respect to the learned model.

Layers         perceptive network




Texture classification using windowed Fourier filters, R. Azencott and J. P. Wang and L. Younes, Pattern Analysis and Machine Intelligence, IEEE Transactions on  19  148--153  (1997)

This paper proposes to represent texture using the energy (sum of squares) of Gabor transforms over small windows. When textures are modeled as stationary Gaussian random fields, these features can be interpreted as non-parametric estimators of the spectral density. This leads to the definition of a distance between textures based on symmetrized Kullback-Leibler distances, between stationary GRFs, which take a very simple form in terms of spectral densities.



Synchronous Random Fields provide
a representation for random fields over discrete grids that can be sampled from using massively parallel schemes that update all variables at the same time. The following papers introduce and study these models in the context of Image Processing and Neural Networks.

Synchronous Boltzmann machines and curve identification tasks, R. Azencott and A. Doutriaux and L. Younes, Network: Computation in Neural Systems  4  461--480  (1993)
Synchronous random fields and image restoration, L Younes, Pattern Analysis and Machine Intelligence, IEEE Transactions on 20 (4), 380-390
Synchronous Boltzmann machines can be universal approximators, L. Younes, Applied Mathematics Letters  9  109--113  (1996)
Representation of Gibbs fields with Synchronous Random Fields., L. Younes, Markov Processes and Related Fields, vol. 2, 285–316.  (1996)

Synchronous image restoration, L Younes, Computer Vision—ECCV'94, 213-217 (1994)
Learning algorithms for extended models of Boltzmann machines, L Younes, ICPR, 602-602, 1994


A three tiered approach for articulated object action modeling and recognition, Le Lu, Gregory D Hager, Laurent Younes, NIPS (2004)


Clutter invariant ATR, D. Bitouk and M. I. Miller and L. Younes, Pattern Analysis and Machine Intelligence, IEEE Transactions on  27  817--821  (2005)
Asymptotic performance analysis for object recognition in clutter, D. Bitouk and M. I. Miller and L. Younes, Proceedings of SPIE  5094  101--108  (2003)
Empirically generated metric spaces for ATR in clutter, D. Bitouk and M. Miller and L. Younes, Signals, Systems and Computers, 2002. Conference Record of the Thirty-Sixth Asilomar Conference on  2  1407--1410  (2002)



An energy minimization method for matching and comparing structured object representations, R. Azencott and L. Younes, Energy Minimization Methods in Computer Vision and Pattern Recognition    441--456  (1997)